| [ | |
| { | |
| "question": "Which organizations from the private sector and civil society were involved in meetings with OSTP regarding the development of the Blueprint for an AI Bill of Rights?", | |
| "contexts": [ | |
| "APPENDIX\n\u2022 OSTP conducted meetings with a variety of stakeholders in the private sector and civil society. Some of these\nmeetings were specifically focused on providing ideas related to the development of the Blueprint for an AI\nBill of Rights while others provided useful general context on the positive use cases, potential harms, and/or\noversight possibilities for these technologies. Participants in these conversations from the private sector and\ncivil society included:\nAdobe \nAmerican Civil Liberties Union (ACLU) The Aspen Commission on Information Disorder The Awood Center The Australian Human Rights Commission Biometrics Institute The Brookings Institute BSA | The Software Alliance Cantellus Group Center for American Progress Center for Democracy and Technology Center on Privacy and Technology at Georgetown Law Christiana Care Color of Change Coworker Data Robot Data Trust Alliance Data and Society Research Institute Deepmind EdSAFE AI Alliance Electronic Privacy Information Center (EPIC) Encode Justice Equal AI Google Hitachi's AI Policy Committee The Innocence Project Institute of Electrical and Electronics Engineers (IEEE) Intuit Lawyers Committee for Civil Rights Under Law Legal Aid Society The Leadership Conference on Civil and Human Rights Meta Microsoft The MIT AI Policy Forum Movement Alliance Project The National Association of Criminal Defense Lawyers O\u2019Neil Risk Consulting & Algorithmic Auditing The Partnership on AI Pinterest The Plaintext Group pymetrics SAP The Security Industry Association Software and Information Industry Association (SIIA) Special Competitive Studies Project Thorn United for Respect University of California at Berkeley Citris Policy Lab University of California at Berkeley Labor Center Unfinished/Project Liberty Upturn US Chamber of Commerce US Chamber of Commerce Technology Engagement Center \nA.I. Working Group\nVibrent HealthWarehouse Worker ResourceCenterWaymap\n62" | |
| ], | |
| "ground_truth": "OSTP conducted meetings with a variety of stakeholders in the private sector and civil society, including organizations such as Adobe, ACLU, The Aspen Commission on Information Disorder, The Australian Human Rights Commission, Biometrics Institute, The Brookings Institute, BSA | The Software Alliance, Cantellus Group, Center for American Progress, Center for Democracy and Technology, Center on Privacy and Technology at Georgetown Law, Christiana Care, Color of Change, Coworker, Data Robot, Data Trust Alliance, Data and Society Research Institute, Deepmind, EdSAFE AI Alliance, Electronic Privacy Information Center (EPIC), Encode Justice, Equal AI, Google, Hitachi's AI Policy Committee, The Innocence Project, Institute of Electrical and Electronics Engineers (IEEE), Intuit, Lawyers Committee for Civil Rights Under Law, Legal Aid Society, The Leadership Conference on Civil and Human Rights, Meta, Microsoft, The MIT AI Policy Forum, Movement Alliance Project, The National Association of Criminal Defense Lawyers, O'Neil Risk Consulting & Algorithmic Auditing, The Partnership on AI, Pinterest, The Plaintext Group, pymetrics, SAP, The Security Industry Association, Software and Information Industry Association (SIIA), Special Competitive Studies Project, Thorn, United for Respect, University of California at Berkeley Citris Policy Lab, University of California at Berkeley Labor Center, Unfinished/Project Liberty, Upturn, US Chamber of Commerce, US Chamber of Commerce Technology Engagement Center, A.I. Working Group, Vibrent Health, Warehouse Worker Resource Center, Waymap.", | |
| "evolution_type": "simple", | |
| "metadata": [ | |
| { | |
| "source": "resources\\Blueprint-for-an-AI-Bill-of-Rights.pdf", | |
| "page": 61, | |
| "filename": "resources\\Blueprint-for-an-AI-Bill-of-Rights.pdf" | |
| } | |
| ], | |
| "episode_done": true | |
| }, | |
| { | |
| "question": "How can statistical biases related to GAI content provenance be assessed and managed in evaluations involving human subjects?", | |
| "contexts": [ | |
| " \n30 MEASURE 2.2: Evaluations involving human subjects meet applicable requirements (including human subject protection) and are \nrepresentative of the relevant population. \nAction ID Suggested Action GAI Risks \nMS-2.2-001 Assess and manage statistical biases related to GAI content provenance through \ntechniques such as re -sampling, re -weighting, or adversarial training. Information Integrity ; Information \nSecurity ; Harmful Bias and \nHomogenization \nMS-2.2-002 Document how content provenance data is tracked and how that data interact s \nwith privacy and security . Consider : Anonymiz ing data to protect the privacy of \nhuman subjects; Leverag ing privacy output \ufb01lters; Remov ing any personally \nidenti\ufb01able information (PII) to prevent potential harm or misuse. Data Privacy ; Human AI \nCon\ufb01guration; Information \nIntegrity ; Information Security ; \nDangerous , Violent, or Hateful \nContent \nMS-2.2-0 03 Provide human subjects with options to withdraw participation or revoke their \nconsent for present or future use of their data in GAI applications . Data Privacy ; Human -AI \nCon\ufb01guration; Information \nIntegrity \nMS-2.2-0 04 Use techniques such as anonymization , di\ufb00erential privacy or other privacy -\nenhancing technologies to minimize the risks associated with linking AI -generated \ncontent back to individual human subjects. Data Privacy ; Human -AI \nCon\ufb01guration \nAI Actor Tasks: AI Development, Human Factors, TEVV \n \nMEASURE 2.3: AI system performance or assurance criteria are measured qualitatively or quantitatively and demonstrated for \nconditions similar to deployment setting(s). Measures are documented. \nAction ID Suggested Action GAI Risks \nMS-2.3-001 Consider baseline model performance on suites of benchmarks when selecting a \nmodel for \ufb01ne tuning or enhancement with retrieval- augmented generation. Information Security ; \nConfabulation \nMS-2.3-002 Evaluate claims of model capabilities using empirically validated methods. Confabulation ; Information \nSecurity \nMS-2.3-003 Share results of pre -deployment testing with relevant GAI Actors , such as those \nwith system release approval authority. Human -AI Con\ufb01guration " | |
| ], | |
| "ground_truth": "Assess and manage statistical biases related to GAI content provenance in evaluations involving human subjects through techniques such as re-sampling, re-weighting, or adversarial training. It is important to ensure that the evaluations are representative of the relevant population and meet applicable requirements, including human subject protection.", | |
| "evolution_type": "simple", | |
| "metadata": [ | |
| { | |
| "source": "resources\\NIST.AI.600-1.pdf", | |
| "page": 33, | |
| "filename": "resources\\NIST.AI.600-1.pdf" | |
| } | |
| ], | |
| "episode_done": true | |
| }, | |
| { | |
| "question": "How can organizations use public feedback to enhance content provenance for GAI models?", | |
| "contexts": [ | |
| " \n52 \u2022 Monitoring system capabilities and limitations in deployment through rigorous TEVV processes; \n\u2022 Evaluati ng how humans engage, interact with, or adapt to GAI content (especially in decision \nmaking tasks informed by GAI content), and how they react to applied provenance techniques \nsuch as overt disclosures. \nOrganizations can document and delineate GAI system objectives and limitations to identify gaps where provenance data may be most useful. For instance, GAI systems used for content creation may require \nrobust watermarking techniques and corresponding detectors to identify the source of content or \nmetadata recording techniques and metadata management tools and repositories to trace content \norigins and modi\ufb01cations. Further narrowing of GAI task de\ufb01nitions to include provenance data can enable organizations to maximize the utility of provenance data and risk management e\ufb00orts. \nA.1.7. \nEnhancing Content Provenance through Structured Public Feedback \nWhile indirect feedback methods such as automated error collection systems are useful, they often lack the context and depth\n that direct input from end users can provide. Organizations can leverage feedback \napproaches described in the Pre-Deployment Testing section to capture input from external sources such \nas through AI red- teaming. \nIntegrating pre - and post -deployment external feedback into the monitoring process for GAI models and \ncorresponding applications can help enhance awareness of performance changes and mitigate potential \nrisks and harms from outputs . There are many ways to capture and make use of user feedback \u2013 before \nand after GAI systems and digital content transparency approaches are deployed \u2013 to gain insights about \nauthentication e\ufb03cacy and vulnerabilities, impacts of adversarial threats on techniques , and unintended \nconsequences resulting from the utilization of content provenance approaches on users and \ncommunities . Furthermore, organizations can track and document the provenance of datasets to identify \ninstances in which AI -generated data is a potential root cause of performance issues with the GAI \nsystem. \nA.1.8. Incident Disclosure \nOverview \nAI incidents can be de\ufb01ned as an \u201c event, circumstance, or series of events where the development, use, \nor malfunction of one or more AI systems directly or indirectly contribute s to one of the following harms: \ninjury or harm to the health of a person or group s of people (including psychological harms and harms to \nmental health); disruption of the management and operation of critical infrastructure; violations of \nhuman rights or a breach of obligations under applicable law intended to protect fundamental, labor, \nand intellectual property rights ; or harm to property, communities, or the environment. \u201d AI incidents can \noccur in the aggregate (i.e., for systemic discrimination) or acutely (i.e., for one individual). \nState of AI Incident Tracking and Disclosure \nFormal channels do not currently exist to report and document AI incidents. However, a number of publicly available databases\n have been created to document their occurrence. These reporting channels \nmake decisions on an ad hoc basis about what kinds of incidents to track. Some, for example, track by \namount of media coverage . " | |
| ], | |
| "ground_truth": "Organizations can enhance content provenance for GAI models by leveraging structured public feedback. This feedback can provide valuable context and depth that automated error collection systems may lack. By integrating pre- and post-deployment external feedback into the monitoring process, organizations can improve awareness of performance changes and mitigate potential risks and harms from GAI model outputs. User feedback can also help in understanding authentication efficacy, vulnerabilities, impacts of adversarial threats, and unintended consequences of content provenance approaches on users and communities. Additionally, tracking and documenting the provenance of datasets can help identify instances where AI-generated data may be causing performance issues in the GAI system.", | |
| "evolution_type": "reasoning", | |
| "metadata": [ | |
| { | |
| "source": "resources\\NIST.AI.600-1.pdf", | |
| "page": 55, | |
| "filename": "resources\\NIST.AI.600-1.pdf" | |
| } | |
| ], | |
| "episode_done": true | |
| }, | |
| { | |
| "question": "How do feedback exercises and red-teaming help combat bias in GAI systems?", | |
| "contexts": [ | |
| " \n29 MS-1.1-006 Implement continuous monitoring of GAI system impacts to identify whether GAI \noutputs are equitable across various sub- populations. Seek active and direct \nfeedback from a\ufb00ected communities via structured feedback mechanisms or red -\nteaming to monitor and improve outputs. Harmful Bias and Homogenization \nMS-1.1-007 Evaluate the quality and integrity of data used in training and the provenance of \nAI-generated content , for example by e mploying techniques like chaos \nengineering and seeking stakeholder feedback. Information Integrity \nMS-1.1-008 De\ufb01ne use cases, contexts of use, capabilities, and negative impacts where \nstructured human feedback exercises, e.g., GAI red- teaming, would be most \nbene\ufb01cial for GAI risk measurement and management based on the context of \nuse. Harmful Bias and \nHomogenization ; CBRN \nInformation or Capabilities \nMS-1.1-0 09 Track and document risks or opportunities related to all GAI risks that cannot be \nmeasured quantitatively, including explanations as to why some risks cannot be \nmeasured (e.g., due to technological limitations, resource constraints, or trustworthy considerations). Include unmeasured risks in marginal risks. Information Integrity \nAI Actor Tasks: AI Development, Domain Experts, TEVV \n \nMEASURE 1.3: Internal experts who did not serve as front -line developers for the system and/or independent assessors are \ninvolved in regular assessments and updates. Domain experts, users, AI Actors external to the team that developed or deployed the \nAI system, and a\ufb00ected communities are consulted in support of assessments as necessary per organizational risk tolerance . \nAction ID Suggested Action GAI Risks \nMS-1.3-001 De\ufb01ne relevant groups of interest (e.g., demographic groups, subject matter \nexperts, experience with GAI technology) within the context of use as part of \nplans for gathering structured public feedback. Human -AI Con\ufb01guration ; Harmful \nBias and Homogenization ; CBRN \nInformation or Capabilities \nMS-1.3-002 Engage in internal and external evaluations , GAI red -teaming, impact \nassessments, or other structured human feedback exercises in consultation \nwith representative AI Actors with expertise and familiarity in the context of \nuse, and/or who are representative of the populations associated with the context of use. Human -AI Con\ufb01guration ; Harmful \nBias and Homogenization ; CBRN \nInformation or Capabilities \nMS-1.3-0 03 Verify those conducting structured human feedback exercises are not directly \ninvolved in system development tasks for the same GAI model. Human -AI Con\ufb01guration ; Data \nPrivacy \nAI Actor Tasks: AI Deployment, AI Development, AI Impact Assessment, A\ufb00ected Individuals and Communities, Domain Experts, \nEnd-Users, Operation and Monitoring, TEVV \n " | |
| ], | |
| "ground_truth": "Feedback exercises and red-teaming help combat bias in GAI systems by implementing continuous monitoring to identify inequities in system outputs across different sub-populations. By seeking direct feedback from affected communities and using structured mechanisms, biases can be detected and addressed. Additionally, evaluating the quality and integrity of data used in training, employing techniques like chaos engineering, and seeking stakeholder feedback can help improve the fairness of AI-generated content. Defining use cases, contexts of use, capabilities, and negative impacts where structured human feedback exercises like red-teaming would be beneficial can also aid in measuring and managing GAI risks related to bias and homogenization. Tracking and documenting risks or opportunities that cannot be quantitatively measured, including explanations for why some risks cannot be measured, is essential for ensuring information integrity in GAI systems.", | |
| "evolution_type": "multi_context", | |
| "metadata": [ | |
| { | |
| "source": "resources\\NIST.AI.600-1.pdf", | |
| "page": 32, | |
| "filename": "resources\\NIST.AI.600-1.pdf" | |
| } | |
| ], | |
| "episode_done": true | |
| } | |
| ] |